Real-time Bayesian optimization of deep brain stimulation for personalized cognitive control enhancement

Identifying effective deep brain stimulation (DBS) parameters for psychiatric disorders has historically been a time-consuming and error prone process due to a lack of an objective and rapid readout of target circuit engagement. Cognitive control may have use as a biomarker of treatment efficacy but it has yet to be shown that DBS parameters can be reliably optimized to produce cognitive control improvements in individual subjects.

We sought to leverage a rat model of DBS-driven cognitive control improvements to determine whether state of the art optimization algorithms could consistently identify effective stimulation amplitudes to enhance cognition.

We delivered periods of active and inactive DBS-like stimulation at variable parameters while rats performed a Set-Shifting task that we previously showed to be stimulation-sensitive. We tested both predefined settings of interest and settings that were personalized to individual animals using Bayesian Optimization. Measurements of task performance including reaction time and accuracy were compared between acute, optimized, and traditional settings to evaluate effects on cognitive control.

Acute stimulation reduced reaction times without hindering accuracy (N=15), replicating the effects previously observed with chronic stimulation. In a second cohort (N=6), optimization of stimulation amplitude successfully reduced reaction times in all animals with comparable effect size to historically best settings.

These findings confirm that optimization techniques can be effective for improving symptomatically-relevant cognitive markers supporting the feasibility of personalized, quantitatively-informed approaches to neuromodulation and target engagement for psychiatric and/or cognitive disorders.